Network-constrained 1st-Order Spatial Point Pattern Analysis

Packages

Code

pacman::p_load(sf,tidyverse,spNetwork,tmap,classInt, viridis)

Import data

Prototype was done focusing on provinces of Thailand. Yet in the actual shiny app due to technical restrictions, we changed to be targeting at Bangkok districts only.

Code

crime_data <- read_rds("../data/rds/accidents_thai.rds") %>% 
  st_transform(crs = 32648)
thai_bound <- st_read(dsn = "../data/thai_adm_boundary",
                      layer = "tha_admbnda_adm2_rtsd_20220121") %>% 
  st_transform(crs = 32648)

Reading layer `tha_admbnda_adm2_rtsd_20220121' from data source 
  `/Users/tangtang/Desktop/IS415 Geospatial Analytics and Applications/project/gga-ThaiRoad/data/thai_adm_boundary' 
  using driver `ESRI Shapefile'
Simple feature collection with 928 features and 19 fields
Geometry type: MULTIPOLYGON
Dimension:     XY
Bounding box:  xmin: 97.34336 ymin: 5.613038 xmax: 105.637 ymax: 20.46507
Geodetic CRS:  WGS 84

Code

thai_road <- read_rds("../data/rds/thai_roads.rds") %>% 
  st_transform(crs = 32648)

Visualising difference in network constrained kernel density through different filters by columns

Some potential filtering to apply to dataset to allow user to zoom in into specific area he/she is interested in:

• Accident categories

• Fatal accident [yes/no]

• Province (MUST-HAVE)

But let’s visualised the overall look

Code

plot(crime_data)

Code

table(crime_data$fatal_accident)

  no  yes 
5974  107 

Code

table(crime_data$accident_categories)

    driver_factors   external_factors             others           speeding 
               413                167                506               4857 
traffic_violations 
               138 

Code

table(crime_data$province_en)

< table of extent 0 >

Example: filter by province = “Bangkok”, accident_categories = “speeding”, fatal_accident == “yes” (Allowing users to change freely)

Code

province_factor = "Bangkok"
accident_category = "speeding"
fatal = "yes"

crime_data_filtered = crime_data %>% 
  filter(accident_categories == accident_category,
         fatal_accident == fatal)

Network-constrained Kernel Density Estimation

Get the portion of road network for the selected province

Code

thai_prov <- thai_bound %>% 
  filter(ADM1_EN == province_factor)

Code

tmap_mode("view")
tm_shape(thai_prov) +
  tm_polygons() +
tm_shape(crime_data_filtered) +
  tm_dots() +
tm_shape(thai_road) +
  tm_lines()

Code

thai_road <- thai_road %>%
  filter(fclass %in% c("motorway","primary","secondary","tertiary"))
road_prov <- st_intersection(thai_road, thai_prov)
road_prov_sf <- st_cast(st_collection_extract(road_prov,
                                             type = "LINESTRING"),"LINESTRING")

Prepare lixel object

1st try with 1000m diameter

Code

lixels <- lixelize_lines(road_prov_sf,
                         1000,
                         mindist = 500)

Generate line centre points

Code

samples <- lines_center(lixels)
densities <- nkde(road_prov_sf,
                  events = crime_data_filtered,
                  w = rep(1, nrow(crime_data_filtered)),
                  samples = samples,
                  ## important
                  kernel_name = "quartic",
                  bw = 500,
                  ## important
                  div = "bw",
                  method = "simple",
                  digits = 1,
                  tol = 1,
                  grid_shape = c(1,1),
                  max_depth = 8,
                  # aggregate events within 10m radius (faster calculation)
                  agg = 10,
                  sparse = TRUE,
                  verbose = FALSE
                  )

Generate plot

Code

samples$density <- densities
lixels$density <- densities

samples$density <- samples$density*1000
lixels$density <- lixels$density*1000

Code

tmap_mode('view')
tm_shape(lixels) +
  tm_lines(col="density",
           lwd = 3)

Code

tmap_mode('plot')

UI design for Analysis

Storyboard of NKDE

The user would be able to filter the dataset to see impact different categories on the network-constrained kernel density map created and use the interpretation rules given to derive at a conclusion.